evaluation-task · Source-linked discovery

IFEval: Instruction-Following Evaluation

Evaluates how well language models can strictly follow detailed instructions, such as writing responses with specific word counts or including required keywords.

Open interactive record →
OriginJeffrey Zhou, Tianjian Lu, Swaroop Mishra et al.TopicsGeneral capabilityStatusimported

Can support

Not independently assessed by FronteraEval yet.

Cannot support by itself

No inference beyond the upstream source should be made until the protocol is reviewed.

Original sources